oldphotocolor // neural photo colouriser, on your device

drop an old black-and-white photo — a real colour model runs in your browser and adds natural colour. compare with the before/after slider, then download. nothing uploaded, no signup.

My grandmother's photos are all black and white, and the paid "colorize" sites wanted a subscription and an upload of pictures of my family to some cloud I'd never heard of. A colouriser is a neural model, sure — but browsers can run neural models now. So this one loads a real colourisation network (DDColor) into the tab, adds colour on my own laptop, and never sends the photo anywhere. First run downloads the model once; after that it works on a plane. The colours it guessed for my grandmother's garden were close enough to make my mother cry.

— for a shoebox of old photographs

What it does with a photo

  1. Runs a real model locally. DDColor, a neural colouriser, executes in your browser via WebAssembly — no upload.
  2. Predicts natural colour. It infers likely colours from the content — skin, sky, foliage, wood — not a flat tint.
  3. Keeps your detail. Colour is recombined with the original full-resolution brightness, so sharpness is preserved.
  4. Compare & download. Drag the before/after slider, then save a PNG. No account, no watermark.

Why oldphotocolor

Worth knowing

Are the colours guaranteed correct?

No — they're plausible guesses. The model can't know a specific dress was red; it infers likely colours from what it learned. Skin, sky and greenery usually convince; unusual objects are its best guess.

Why the first download?

The standard model is 62 MB — an int8-compressed edition of the network that colours all but the trickiest photos identically to the original. It downloads once, then caches for instant, offline use. Want the untouched network? Flip the model control to maximum fidelity (226 MB) and compare for yourself.

My photo is large.

Very large images are worked at up to 1600px for speed and memory, then colour is mapped back. The output stays sharp because it uses your original brightness.

Where does the photo go?

Nowhere — it's processed in the tab and never transmitted. Close the tab and it's gone.

Darkroom notes

How photo colourisation worksBrightness stays, colour is predicted — why a network only has to guess two channels. Getting the best from old photosScanning, contrast and expectations — how to help the model do its best work. Colourise without uploading your photosFamily archives are private. Why a real model can now run in the browser.